The Power and Pitfalls of Omics: George Davey Smith’s storming talk at ME/CFS conference
Read about the talk that stole the show at a recent ME/CFS conference in Simon McGrath's two-part blog.
Discuss the article on the Forums.

Machine Learning-assisted Research on CFS

Discussion in 'General ME/CFS Discussion' started by mariovitali, May 8, 2017.

  1. mariovitali

    mariovitali Senior Member

    Messages:
    850
    Likes:
    716
    @Hip

    CC @Jesse2233

    Please read the following e-mail i received from a Professor who i will not disclose :


    I will not get into the argument on whether this Professor is a true Scientist or not. But i cannot but commend the attitude where we have on one hand Specific concerns being presented that require attention but on the other hand not readily dismissing a new technology.

    I do understand what Professor Edwards meant but there is also a possibility that given the amount of information, such bias gets "cancelled out" so to speak.

    But please, allow me to elaborate further : Given the fact that there is a lot of garbage out there (but also non-garbage) we have a set of Hypotheses being created using some Analytical techniques.

    I have repeatedly asked Proffessor Edwards to comment on the output (=Hypotheses) that originated from these techniques and whether the hypotheses being generated could be valid. He never replied despite my repeated questions. If you could comment on why this is so (and even better him doing so) it would be great. For all of us.

    I am providing some links to these dialogs :

    http://forums.phoenixrising.me/inde...sted-research-on-cfs.51283/page-4#post-853658

    and

    http://forums.phoenixrising.me/inde...sted-research-on-cfs.51283/page-4#post-853607

    and

    http://forums.phoenixrising.me/inde...sted-research-on-cfs.51283/page-4#post-853727
     
    Last edited: Oct 17, 2017
  2. Hip

    Hip Senior Member

    Messages:
    9,481
    Likes:
    15,009
    Can you discuss those hypothesis-creating techniques, or are they your own proprietary algorithms that you wish to keep secret?

    If I were writing your software algorithm, I would run the algorithm on diseases that we have recently discovered the cause of, but restrict the algorithm to papers published before the cause was known. Then you could see how many times your algorithm guesses right. In other words, run your algorithm on historical data that we already known the answer to, and see if your algorithm comes up with the right answer. That would be impressive if it could.

    I wrote some predictive software many years ago, and I tested out in this way, using historical data that we know the results of: in my MSc in cognitive science, in my project I was using computer neural networks to model brain function. Just for fun, I decided to write a neural network-based prediction algorithm for the football pools (the lottery based on guessing the results of UK football matches). I trained my neural network using past football match data, and then I tested its ability to of my neural network to predict some further past football games (which the network had never seen). I found, incidentally, that my neural network was able to predict the correct football result 10 times better than chance alone, but this was not sufficient to stand a reasonable chance of winning the football pools.

    So you could the same thing, and test your algorithm on historical data.

    I find the idea of your algorithm very interesting; but without some demonstration of its abilities, using historical data in the way I described, it remains an unproven idea.



    It takes many months (if not years) of effort and brain powder to understand the connection that any given pathway may or may not have to ME/CFS (or any other disease for that matter). Most medical researchers have specialized areas of biomedicine that they know a lot about, and they usually also have a broad, but less in-depth, understanding of biomedicine in general.

    So what you are asking is for someone to drop their current activities, and devote several months to investigating some pathway that your algorithm has suggested. And without any evidence that it is the right answer. You might find some researchers willing to do that, but I expect most will not, as they have their own interests.

    Especially as your algorithm provides no context. For example, you say your algorithm picked up on pyruvate dehydrogenase before Fluge and Mella discovered a possible dysfunction in PDH. But with Fluge and Mella's investigations, we see the entire context in which PDH fits in, so you can understand the whole picture.

    But if your algorithm just points to PDH without any context, it's hard to know what to make of it. If your algorithm were to throw out say "alpha lipoic acid", there are thousands of pathways that might potentially be involved; without providing any further context, how is a researcher going to begin to guess how ALA might be linked to ME/CFS?

    If you were able to extract a bit of context from your algorithm, that would help I think. I don't know how your algorithm works, so don't know if providing context would be possible, but it would be helpful. As would validating your algorithm on diseases with known causes.
     
    Last edited: Oct 17, 2017
  3. mariovitali

    mariovitali Senior Member

    Messages:
    850
    Likes:
    716
    @Hip

    See my comments inline :
    I think we must first define what is "hypothesis" here. The system extracts Topics that are most likely related to a certain condition. It does not generate a complete hypothesis in the way that -if my understanding is right- you describe. As you also noted, the system does not provide context. In other words we do not know if Tocotrienol is good or bad for us. However in our case, the system identifies several Topics that are part of specific pathways (e.g Bile Acids, Glucolysis). Both of these are related to the Liver so this is a Hypothesis being generated based on these Topics.


    I agree with this practice. However, please have in mind that i was ill and developing this system at the same time. My health was deteriorating very rapidly. I was found being in a pre-diabetic state, Hypothyroid and with many other issues. You are talking about "backtesting" the system which is a standard practice in Machine Learning. I also feel that the rate with which the system puts pieces together is not linear (if this makes sense to you) so going back 10 years may affect the system's performance way much more than going back 2 years.

    This is not the case. I presented to Professor Edwards a series of Papers pertinent to MERTK , GAS6 and some other Genes related to Vitamin K metabolism. If my memory serves me well, back then the system was looking at almost 600 Topics, it selected 4 of them and all 4 had functions related to autoimmune disease. The comment i asked by Professor Edwards was for this fact, i did not ask him to assess the whole Theory.

    I think that the system not only described PDH before Fluge/Mella but also Phospholipids and Bile Acids as found by Naviaux. I must also see if PPARγ importance was found before Fluge/Mela. Regarding the lack of context you describe, this is correct. The system is meant to be used along with Researchers who have the responsibility to generate a complete Hypothesis. But imagine how helpful a system would be that gives you a guidance and enables you to focus on the root of the problem (hopefully) and not to an area that has nothing to with the Research Topic at hand.
     
  4. mariovitali

    mariovitali Senior Member

    Messages:
    850
    Likes:
    716
    Here is a quick update and the latest Network Analysis graph:


    Screen Shot 2017-11-08 at 21.39.22.png



    We take note of the following :


    -The important roles of NAD and Tocotrienol (it is not suggested to take supplemental Tocotrienol)

    -How glycosyltransferases connect Endoplasmic Reticulum Stress/ Unfolded Protein Response topics (under Green rectangle) and more central topics such as NAD and Tocotrienol.

    -We note also the role of Glycoproteins (Red rectangle) with Topics related to Inflammatory Processes.

    -There is also a part on the Network (not shown) which has many Liver-related Topics.

    Glycosyltransferases may be a possible marker for ME/CFS.

    Apart from Network Analysis, there is an increasing ranking in the importance of GRB2 and SH2B3 (more coming soon regarding these two genes)
     
  5. aquariusgirl

    aquariusgirl Senior Member

    Messages:
    1,426
    Likes:
    440
    Mario: the moderators @ Low Oxolates group recognize the syndrome comes w/gallbladder problems.

    But they haven’t figured out the mechanism.
     
    mariovitali likes this.
  6. mariovitali

    mariovitali Senior Member

    Messages:
    850
    Likes:
    716
    @Violeta

    Can you provide a link please?
     
  7. aquariusgirl

    aquariusgirl Senior Member

    Messages:
    1,426
    Likes:
    440
    @mariovitali .'.Trying Low Oxolates is a closed facebook group so not sure I can. In her post on iron , Alethea suggest that B1 uses up manganese. Maybe there is a connection there?
     
    Last edited: Nov 12, 2017
  8. aquariusgirl

    aquariusgirl Senior Member

    Messages:
    1,426
    Likes:
    440
    @mariovitali

    There is a line of research that associates low zinc with various pyschiatric problems. I read that zinc supplementation is being advised for those with treatment resistant depression.

    Ok, so what has this got to do with CFS? Well this paper below suggests there are genetic variants in certain zinc transporters that result in low intracellular zinc even in the presence of normal serum zinc. The zinc/copper ratio is an issue in most neurological disease....and certainly there is a zinc, b6, oxalate connection.

    "Furthermore, inefficient genetic variants in zinc transporter molecules that transport the ion across cellular membranes impede its action even when circulating zinc concentrations is in the normal range."
    " In addition, zinc is an agonist for GPR39 activation and for mTOR (mammalian target of rapamycin) (Szewczyk et al., 2015). "

    https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5492454/

    I also note that one of these zinc transporter genes has been associated with Ehlers Danlos.
    https://www.ncbi.nlm.nih.gov/pubmed/18985159/
     
    Last edited: Nov 12, 2017
  9. mariovitali

    mariovitali Senior Member

    Messages:
    850
    Likes:
    716
    Here is a quick update everyone.


    We have a new recovery :)

    I got the OK from this person to share his real name and DNA Data to the OMF. I sent this information to Janet Dafoe just today.

    This person had mid-to-severe symptoms of ME/CFS for 3 years. At times he had severe symptoms. He followed a personalized regimen for 5 months. What is really interesting with this particular case is that for the past 3 months he receives no treatment and despite this he has no symptoms. Of course we do not know if this will last.

    I believe that giving his data to OMF is imporant, since they may be able to identify differences between patients having remission of symptoms vs patients with no remission.


    Research-wise : we have an increasing importance of Omega-3 fatty acids mainly through the fact that they induce Peroxisome Proliferator PPARγ.

    I will post more when i find some time.
     
    Last edited: Dec 7, 2017 at 6:03 AM

See more popular forum discussions.

Share This Page